ChanServ changed the topic of #dri-devel to: <ajax> nothing involved with X should ever be unable to find a bar
<mattst88>
what service are all of those clients on, that just died?^
dogukan has joined #dri-devel
dogukan has quit []
<daniels>
IPv6
<daniels>
oh, looks like a Matrix bridge
<mattst88>
well, glad I stuck with v4 :)
<mattst88>
ah, makes sense. that's probably what all those [m] mean
<karolherbst>
HdkR: tell that to intel
<karolherbst>
apparently they don't have atomic cas for 64 bit local ops
<karolherbst>
¯\_(ツ)_/¯
<karolherbst>
I saw the lowering code...
<karolherbst>
it's... "special"
<karolherbst>
mattst88: I always thought that means mobile... but given the state of mobile IRC clients, I think matrix is the better explanation
<mattst88>
heh, yeah, I did too
<HdkR>
karolherbst: I understand that nightmare. Do they have a way to make two ops atomic with each other, or will it always have a chance to tear?
ngcortes has joined #dri-devel
jkrzyszt has quit [Ping timeout: 480 seconds]
<karolherbst>
HdkR: it involces reserving space for locks :)
<karolherbst>
*involves
<karolherbst>
they can do 32 bit ones, just not 64 bit
<karolherbst>
I guess there are worst options :D but hey
undvasistas[m] has joined #dri-devel
<karolherbst>
I thought I saw it all, but here it is: "LLVM IR to Bash 4.x Transpiler"
<HdkR>
ah, so all 64bit atomics hit the lock. Painful
<karolherbst>
very
<karolherbst>
HdkR: well.. only on local memory though
<karolherbst>
it's not like NV is a lot better here, but at least they give us a 64 bit CAS
<HdkR>
64bit and 128bit CAS is very nice to have
<karolherbst>
yeah
<HdkR>
Lets you cheat most things
<karolherbst>
well.. perf tanks either way
<HdkR>
Of course. Correctness matters more though
<karolherbst>
or well.. just declare silly features nobody uses as optional and let the app dev rethink their life choices
<HdkR>
Hey, split locks are now an optional feature on the Linux kernel :)
<karolherbst>
lol
Lucretia has quit []
<HdkR>
They happen all.the.time.
<HdkR>
`x86/split lock detection: #AC: CHTTPClientThre/226380 took a split_lock trap at address: 0xea59c3c3`
<karolherbst>
:D
* HdkR
throws Steam under the bus
<karolherbst>
don't use crappy archs then?
<HdkR>
:<
<karolherbst>
but seriously.. whose fault is that? :P
<HdkR>
Intel for supporting it in hardware ;)
<karolherbst>
:D
<karolherbst>
yeah and becayse they suck perf wise, we now have to warn about those and tell devs to please use something else :D
<HdkR>
Or do the AMD thing and get silently broken code
<HdkR>
oop, it tear
<karolherbst>
ohh crap.. memset was this super useless function
pnowack has quit [Quit: pnowack]
<bnieuwenhuizen>
HdkR: sounds like AMD are the only ones supporting 64-bit atomics fully on local memory then it seems
<HdkR>
Neat
<HdkR>
Wonder if that is mainly for the compute cards
<bnieuwenhuizen>
hmm, and for a prize of weird instructions "DS_WRXCHG2ST64_RTN_B64": Write-exchange 2 qwords with a stride of 64 qwords.
<bnieuwenhuizen>
wonder what use they found for that one
<karolherbst>
not a stride of 64 bit?
<bnieuwenhuizen>
nope
<karolherbst>
strange
<HdkR>
Almost seems like something used for a wave wide reduction using cas
<bnieuwenhuizen>
oh my guess is better use of the offset encoding space allowing the entries to be more spaced out of they are aligned. There are read2/write2 ops too
<HdkR>
oh, fun
ngcortes has quit [Remote host closed the connection]
jewins1 has quit [Ping timeout: 480 seconds]
khfeng has joined #dri-devel
<mareko>
I thought everybody had 64-bit shared memory atomics
sdutt has quit []
sdutt has joined #dri-devel
_alice has joined #dri-devel
Peste_Bubonica has quit [Quit: Leaving]
boistordu has joined #dri-devel
boistordu_ex has quit [Ping timeout: 480 seconds]
MatrixTravelerbot[m] has joined #dri-devel
Company has quit [Quit: Leaving]
mbrost_ has quit [Read error: Connection reset by peer]
ppascher has joined #dri-devel
Eighth_Doctor has joined #dri-devel
cwfitzgerald[m] has joined #dri-devel
jhli has quit [Ping timeout: 480 seconds]
slattann has joined #dri-devel
user1tt[m] has joined #dri-devel
pmoreau has joined #dri-devel
pmoreau is now known as Guest4069
mattrope has quit [Remote host closed the connection]
JohnnyonFlame has quit [Read error: Connection reset by peer]
Duke`` has joined #dri-devel
tzimmermann has joined #dri-devel
mlankhorst has joined #dri-devel
gnustomp[m] has joined #dri-devel
slattann has quit []
gpoo has quit [Ping timeout: 480 seconds]
YuGiOhJCJ has joined #dri-devel
danvet has joined #dri-devel
pochu has quit [Ping timeout: 480 seconds]
slattann has joined #dri-devel
Duke`` has quit [Ping timeout: 480 seconds]
jkrzyszt has joined #dri-devel
pcercuei has joined #dri-devel
exit70[m] has joined #dri-devel
cmarcelo has joined #dri-devel
rasterman has joined #dri-devel
sdutt has quit [Ping timeout: 480 seconds]
vivekk has quit [Ping timeout: 480 seconds]
RAOFhehis[m] has joined #dri-devel
<mareko>
what is the deal with tegra in mesa? do we really need to have an in-tree shim?
pochu has joined #dri-devel
<HdkR>
This was before kmsro, so it just hasn't been switched over to that at this point?
<alyssa>
anholt_: It looks good to my untrained eyes, but I don't claim familiarity with glsl->nir stuff
YuGiOhJCJ has quit [Quit: YuGiOhJCJ]
Strit[m] has joined #dri-devel
<emersion>
leandrohrb2: do you want to add the new R-b and A-b tags to that vulkan MR?
shfil has joined #dri-devel
aura[m] has joined #dri-devel
bluebugs has quit [Quit: Leaving]
ServerStatsDiscoverertraveler4 has joined #dri-devel
<jekstrand>
lynxeye: Thanks!
cedric has joined #dri-devel
cedric is now known as bluebugs
<alyssa>
jekstrand: Do we have a common code definition of a colour union
<jekstrand>
alyssa: Uh... We've got many common things?
<jekstrand>
alyssa: No, not really.
<alyssa>
pipe_color_union and VkClearColorValue have identical definitions but different names so can't share clear colour packing
<alyssa>
and adding a 3rd definition (pan_color_union) and casting seems like a worst-of-all-worlds
<alyssa>
pinging you since I assume isl deals with this
<alyssa>
oh... isl_color_value...
<jekstrand>
Yup. :D
<alyssa>
which is in turn identical, ugh
<alyssa>
Maybe it makes sense to move the gallium one to util/format/ and then switch isl to that?
<jekstrand>
I don't know
<jekstrand>
They're all the same. Cast/memcpy.
<jekstrand>
Or use pipe in your Vulkan driver
<alyssa>
(Context: I found a bug in the panfrost clear colour packing. panvk inherited the same bug.)
<alyssa>
actually like 12 bugs.
<alyssa>
this is r/e code that hasn't been touched in a year. it real bad.
<alyssa>
shocking how nothing in deqp-gles2/3/31 was affected by it :-p
<jekstrand>
:-(
<alyssa>
buggy packing of clear colours with funny formats
* jekstrand
loves asserts, even in the kernel
gouchi has joined #dri-devel
shfil has quit [Ping timeout: 480 seconds]
<alyssa>
jekstrand: WARN_ON?
<jekstrand>
alyssa: Yup
<jekstrand>
Or BUG_ON
<jekstrand>
Though that one has a nasty habbit of taking down CPU cores. :)
<alyssa>
there's a torvalds rant about BUG_ON
<imirkin>
definitely don't use BUG_ON for asserts
<imirkin>
BUG_ON is for when there's absolutely no way to continue
<jekstrand>
I know
<alyssa>
I would link the rant but it probably violates fd.o CoC ;p
<kisak>
alyssa: with your personality, I wouldn't be surprised if you wrote up a rant, then replaced all the expletives with Harry Potter references.
<alyssa>
kisak: My Little Pony, but yes
<kisak>
my bad, My Little Punny works too
<alyssa>
Word play is magic?
<cmarcelo>
pendingchaos: by your last comment in https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9938 not clear you want changes to happen or not. in summary my take there: it is ok to be more conservative wrt removing loops with memory loads -- so I wouldn't add more cases and keep the patch as is.
slattann has joined #dri-devel
<pendingchaos>
cmarcelo: yes
reductum has quit [Ping timeout: 480 seconds]
<cmarcelo>
pendingchaos: could you resolve that comment then? and... do you plan to Rb/Ab? :-)
mbrost has joined #dri-devel
pendingchaos has quit [Quit: No Ping reply in 180 seconds.]
reductum has joined #dri-devel
slattann has quit []
pendingchaos has joined #dri-devel
<alyssa>
kisak: Debugging some stuff rn
<alyssa>
I'm just so.... peeved!
* alyssa
blushes
<alyssa>
..... glClearColor gets dithered doesn't it :V
nirmoy has quit []
pendingchaos has quit [Quit: No Ping reply in 180 seconds.]
gouchi has quit [Remote host closed the connection]
pendingchaos has joined #dri-devel
shfil has joined #dri-devel
<cmarcelo>
pendingchaos: could you resolve that comment then? and... do you plan to Rb/Ab? :-)
<cmarcelo>
thanks
<karolherbst>
alyssa: do you have a M1 macbook actually? I've heard it can only do one external display, but I assume it's just macos being stupid and the hw can drive 2 displays in total, just that the internal needs to be disabled or something, right?
shash_sha has joined #dri-devel
shashank_sharma has quit [Remote host closed the connection]
yogesh_mohan has quit [Quit: WeeChat 2.3]
yogesh_mohan has joined #dri-devel
yogesh_mohan has quit []
<alyssa>
karolherbst: Just a mini
<alyssa>
AFAIU, the SoC has two display controllers and each can only drive a single output
<alyssa>
so mini can do two displays (1 HDMI, 1 displayport)
<alyssa>
the macbooks, I guess one of them is tied up with the internal, dunno if that's tied in hardware/firmware/software
cwfitzgerald[m] is now known as cwfitzgerald
ngcortes has joined #dri-devel
<karolherbst>
alyssa: yeah...
<karolherbst>
not sure
<alyssa>
"Passed 0/52"
<alyssa>
unit testing is going well.
bcarvalho has quit [Read error: Connection reset by peer]
bcarvalho has joined #dri-devel
shashank_sharma has joined #dri-devel
shash_sha has quit [Ping timeout: 480 seconds]
<karolherbst>
yay
sdutt has quit [Ping timeout: 480 seconds]
lemonzest has quit [Quit: Quitting]
tzimmermann has quit [Quit: Leaving]
sneil has joined #dri-devel
bcarvalho has quit [Remote host closed the connection]
bcarvalho has joined #dri-devel
hch12907 has quit [Read error: Connection reset by peer]
<milek7>
Mesa: User error: GL_INVALID_OPERATION in glTexImage2D(format = GL_RED, type = GL_UNSIGNED_BYTE, internalformat = GL_R8UI)
<milek7>
is this not allowed? opengl texture formats are so weird...
Guest4158 is now known as icecream95
<imirkin>
milek7: GL_RED_INTEGER i think
<milek7>
yeah, thanks
rasterman has quit [Quit: Gettin' stinky!]
Duke`` has quit [Ping timeout: 480 seconds]
<alyssa>
Not sure Nanley Chery is here but if so...
<alyssa>
Thank you for listing the failing dEQP cases in your commit message fbbf79377bc ("iris: Remove the CCS_D fallback")
<alyssa>
It tipped me off the bug in Panfrost leading me to fail to the same t
<alyssa>
test cases. Ended up rewriting the whole clear colour packing code, adding a few dozen unit tests, ... a few hours to getting to passing those unit tests later, the dEQP cases are passing too now :-)
shfil has quit [Ping timeout: 480 seconds]
<alyssa>
Didn't investigate too closely where the CTS test was going wrong ... the suspect code turned out to be so deeply and utterly broken that it might not have been just one thing ;-p
<alyssa>
Probably swapping B/R in the clear colour when an sRGB unorm8 surface is used. But that's only one of a massive number of bugs found and fixed...
<karolherbst>
is there a way inside a gallium driver to get _all_ query objects?
<karolherbst>
or would a driver have to track them itself?
<imirkin>
karolherbst: easier to just keep a per-screen generation count
<imirkin>
which you bump whenever you want the query objects to "know" something big has changed
<karolherbst>
ohh
<karolherbst>
good idea actually
<imirkin>
and if you end up with more than 2^32 such events, then you're in trouble.
<imirkin>
;)
<karolherbst>
:D
<jenatali>
Just make it a uint64_t and be done with it
<karolherbst>
we actually have this sequence number anyway
<imirkin>
jenatali: well, i happen to know he's talking about hang detection
vivek has quit [Remote host closed the connection]
<imirkin>
so if you have 2^32 hangs throughout the lifetime of the application ...
vivek has joined #dri-devel
<jenatali>
Ah
<jenatali>
Yeah ok
gouchi has joined #dri-devel
<mattst88>
karolherbst, Sachiel, alyssa: FWIW emailing webmaster@khronos.org was successful. I can now access gerrit
<karolherbst>
yay
<alyssa>
mattst88: Oh, I meant to check that uhh
<alyssa>
I spent all day fixing an "easy" CTS case
danvet has quit [Ping timeout: 480 seconds]
<karolherbst>
alyssa: those are the best
<alyssa>
karolherbst: ended up rewriting a block of totally buggy code and added unit tests for the thing, fixing a lot more bugs than the one thing that case tripped over
jhli has joined #dri-devel
<mattst88>
alyssa: I've learned never to try to predict what is easy and what isn't :)
<jekstrand>
mattst88: I, on the other hand, have simply learned to be ok with my predictions being wrong. :P
jhli has quit [Ping timeout: 480 seconds]
vivek has quit []
gouchi has quit [Remote host closed the connection]
jhli has joined #dri-devel
mlankhorst has quit [Ping timeout: 480 seconds]
<dcbaker>
I have learned to predict that everything is going to be horribly hard, which means that almost everything is surprisingly easy
<karolherbst>
I learned to just do stuff and don't bother with pointless predicitons :p
<Sachiel>
I learned to be more careful when buying advil because some of them have stuff to help you sleep and it's a good idea not to take them during the day when you are trying to work
<HdkR>
I'm filled with undeterred optimism. Maybe that's my problem.
<imirkin>
HdkR: if you were to think that were your problem, your optimism would then be deterred... can't have that
<HdkR>
It's true
<alyssa>
Sachiel: is the only one among us with a good approach to software estimation
bcarvalho has quit [Remote host closed the connection]
alanc has quit [Remote host closed the connection]
alanc has joined #dri-devel
pnowack has quit [Quit: pnowack]
drawat_ has joined #dri-devel
reductum has quit [Ping timeout: 480 seconds]
drawat has quit [Ping timeout: 480 seconds]
jkrzyszt has quit [Ping timeout: 480 seconds]
dogukan has quit [Quit: Konversation terminated!]
bluebugs has quit [Quit: Leaving]
iive has quit []
adjtm has quit [Remote host closed the connection]
adjtm has joined #dri-devel
padovan4 has quit []
padovan4 has joined #dri-devel
<alyssa>
dcbaker: I've been in bug fixing mode. There is a lot of Cc: mesa-stable of big changes. I am sorry in advance.
<alyssa>
if we feel the need to say <21.2 is hosed for panfrost let me know ;-p
<alyssa>
(My personal laptop is still on 20.3 so there's that.. :P)
jkrzyszt has joined #dri-devel
<karolherbst>
alyssa: shame on you
<icecream95>
alyssa: I am also still using the Debian package on speedy..
<alyssa>
karolherbst: for which part, spamming mesa-stable or running mesa-stable/ ;-p
<karolherbst>
running outdated mesa
<alyssa>
eh
<alyssa>
it means that machine only has es3.0 instead of es3.1 but otherwise shrug
<HdkR>
ooo
<alyssa>
HdkR: hm?
<HdkR>
I forgot about ES 3.1 support :D
<alyssa>
HdkR: ES3.1 on G31/G51/G52/G76 is /almost/ conformant
<HdkR>
niiiice
<alyssa>
I built the latest GLES CTS release, and cherry-picked a fix that landed in VK-GL-CTS master in May, and with my fixes stack that's passing all of {dEQP,KHR]-GLES{2,3,31}
<alyssa>
found out about dEQP-EGL yesterday, have just a few tests to fix there
<HdkR>
That's awesome
<alyssa>
and then there's whatever bugs get unsurfaced by glcts running for 2 weeks straight ....
<HdkR>
wow, need some faster boards :P
<alyssa>
Groan
<alyssa>
surfaceless is speedy enough
<alyssa>
but cts-runner doesn't seem to work with surfaceless (?)
<alyssa>
it does work with X, but then... X...
<icecream95>
alyssa: [Running it for two weeks] could find the leak of shader NIR maybe?
<alyssa>
there's only one?
<icecream95>
Also ssa_constants and sysval_to_id are leaked in the Midgard compiler
<icecream95>
(on 32-bit systems)
* alyssa
glances awkwardly at speedy
<alyssa>
maybe wayland cts would be faster? idk
<alyssa>
I wouldn't think the winsys would matter
<HdkR>
winsys is the bain of everyone. Of course it matters :D
<bnieuwenhuizen>
join the vulkan side, where WSI tests take negligible time compared to the rest of CTS
<alyssa>
bnieuwenhuizen: ugh
<alyssa>
what I don't understand is why even dEQP-GLES3 is so much slower with my X build..
<icecream95>
alyssa: Are you using plain Xorg or Xwayland?
<alyssa>
plain Xorg
lynxeye has quit [Quit: Leaving.]
<alyssa>
mostly unsure why surfaceless glcts doesn't work
ngcortes has quit [Remote host closed the connection]